The weakness of future morphemes and epistemicmodals: modal adverbs measuring speaker’s

confidence

Anastasia Giannakidou and Alda MariUniversity of Chicago and Institut Jean Nicod, CNRS

June 14, 2013

Abstract

In recent work, we have defended the thesis that future morphemes in a numberof European languages associate with epistemic modal bases and therefore behave,to all intents and purposes, as epistemic modals (Giannakidou and Mari to appeara,b, 2013, 2013). When this happens, future morphemes, including English, re-ceive non-future, non-predictive readings, akin to must. In this paper, we enhancethe epistemic modal analysis by considering the co-occurrence of future morphemeswith modal adverbs of varying force such as as possibly, probably, definitely, etc. Wenote first that these adverbs can be used with future morphemes in predictive (‘true’future) uses including e.g. Maybe John will be here at 5, as well as non-predictive(non-future) uses such as That will probably be the mailman. All future variantsare non veridical (Giannakidou 1998, 1999), i.e. epistemically weaker than theirrespective non-modalized positive assertions which are veridical. We argue that theadverbs measure the speaker’s confidence that the actual world is among the bestworlds non-predictive use), or the most reasonable futures (true future use) quanti-fied universally by the future modal. The variable force of the adverb (possibility,probability, necessity) reflects higher or lower confidence. This analysis allows anew understanding of the phenomenon of so-called modal concord. In both predic-tive and non-predictive uses, the adverb serves the function of revealing the rolethe speaker plays in making judgments about the actual world. We call this modalmodulation.

1 Introduction: future morphemes and epistemic modal-ity

The question of whether the category ‘future tense’ in natural languages is a tense ormodality has received a lot of attention in linguistic semantics, and both answers havebeen explored (for modal accounts see Bertinetto, 1979; Enc 1996, Copley, 2002; Squartini,2004; Kaufmann 2005, Mari, 2009,to appear Klecha 2011, Giannakidou 2012, Giannakidouand Mari 2012, 2013; for a defense of the temporal analysis see Kissine 2008). Certainly,the English future word ‘will’ is a modal verb, and admits purely epistemic readings. Weread, e.g. in Palmer, that ‘it is tempting to refer to the meaning of will as probability,

1

alongside possibility and necessity for may and must . A better paraphrase is in terms ofconclusion: ‘A reasonable conclusion is that.., (Palmer 1987: 136)’.’

(1) a. The French will be on holiday this week.b. No doubt, you’ll remember John.

The sentences above are conjectural: will conveys epistemic modality: given what I know,it is quite likely or it is quite plausible, or in cases of stronger certainty, it must be thecase that that the French are on holiday this week. Other non-predictive uses of willcirculated in the literature are generic or with mixed ability and epistemic readings:

(2) a. Ed will get upset over nothingb. Ed will lay in bed all day reading trashy novels. (Huddleston 1995)c. As far as I know, oil will float on the water. (Haegeman 1993)

In these cases too, will comes out as the product of reasoning based on knowledge thatthe speaker has (see Giannakidou 2012, Mari, 2009,2010, Chiou 2012, and Tsangalidis 1998for earlier discussion of similar Greek data).

Purely modal uses for futures have also been observed for Dutch. Very much in thespirit of the quote from Palmer, Broekhuis and Verkuyl 2013 propose that Dutch zullen isan epistemic modal operator used to express that the prejacent proposition is the resultof reasoning based on information that judged as ‘reliable and well-founded’. To quotefrom the conclusions:

‘We think that the best way to describe the meaning of WILL(p) is to say that thespeakers using this modal verb feel sufficiently confident to say p is true at n or is tobe made true at i’. This confidence relies on information judged as reliable and well-founded. It may take all sorts of form dependent on the situation: as a hypothesis, aconfident expectation, a reassurance, etc. What these circumscriptions have in commonis that the speaker has entrance to sufficiently many worlds to be able to pick out theones that seem convincing.” (Broekhuis and Verkuyl 2013: Conclusions).

This is a very appealing analysis of the future morpheme, and indeed resonating withrecent proposals we made about Greek and Italian futures (Giannakidou 2012, Mari,2009,2010; Giannakidou and Mari 2012, 2013). The idea that the future is modal isalso common in the Italian literature (Bertinetto 1979, Mari 2012 and more works citedtherein), and modal uses of Greek tha have also been noted in Greek grammars and lit-erature. Morevoer, there is considerable evidence from Slavic languages that future mor-phemes do not have purely temporal contribution either (see e.g. Blaszack and Klimek-Jankowska 2012).These crosslinguistic studies jointly support the idea that if indeed inso many languages future morphemes are systematically employed to convey epistemicreadings, then the ‘future’ as a temporal category becomes redundant. It must be betterunderstood as relying on epistemic modality.

The implications of this idea are large, therefore in the present paper, we would liketo present additional arguments to support the epistemic analysis of the future. We willstudy the co-occurrence of future morphemes with modal adverbs, in producing bothpurely epistemic and predictive ("real" future) readings. We will show first a parallelwith epistemic modals such as must, which seem to be equivalent to the future, butthen we will also discuss possible restrictions. Though the apparent co-occurrence ofadverbs with modals makes us think of ’modal’ concord, we will show that the relationbetween the adverb and the modal is more loose, and goes beyond matching force, whichis usually required for modal concord (Huitink 2012). We will claim that future sentences

2

are perspectival, i.e. they express the speaker’s judgment about the possible truth ofthe prejacent proposition, and non veridical, i.e. they are epistemically weaker thatunmodalized assertions (Giannakidou 1998, 1999, to appear). The function of the adverbis to ’modulate’ the perspective: it measures the speaker’s confidence about where to placethe actual world. The bottomline of the discussion will be that epistemic modal sentences,and therefore also future sentences, are multidimensional, and that modal morphemes,by themselves do not make claims about the actual world. The adverbs are needed to dothat, and this gives us a more refined view of the phenomenon of modal ‘concord’.

The discussion proceeds as follows. In section 2, we discuss the purely epistemic usesof the Greek and Italian morphemes, presenting also the main data from adverbs andthe parallel with necessity modals. It will become evident that the claims with futurecan be relatively weak, as observed especially in Italian. In section 3, we make similarobservations about the predictive use, where we find even weaker adverbs. In section 4,we outline our analysis of FUT as an necessity modal, and discuss the role of the adverbs.We consider possible implementations based on Giannakidou 2002 and Lassiter 2011, butend up suggesting that the adverbs do not compose directly with the modal but are ratherpresupposition inducers. They bring in the speaker’s perspective into the discussion andmodulate the modality. As a result, epistemic modals at a least, end up having a multi-dimensional semantics, with a subjective component that makes reference to speaker’sconfidence about where to place the actual world.

2 Purely epistemic readings in Greek and Italian: non-veridicality and epistemic weakening

For the future, Italian and Greek do not have a modal verb, but employ a morpheme -which we will call FUT in this paper: a bound one (Italian), and a particle that precedesthe tensed verb (Greek). FUT can be followed by a nonpast or a past tense form and retainan epistemic, non-predictive reading (Mari to appear, Giannakidou 2012, Giannakidouand Mari 2012a,b):

(3) a. Ithe

AriadneAriadne

thaFUT

troieat.non-past3sg

tora.now

(epistemic, now)

b. GiacomoGiacomo

oranow

mangeràFUT-eat

’Giacomo/Ariadne must/will be eating now’

(4) Ithe

AriadneAriadne

thaFUT

milisetalk.past.3sg

xthes.yesterday

(epistemic, past)

‘Ariadne must have spoken yesterday.’

(5) GiovanniGiovanni

saràFUT-be

malato.sick

(epistemic, now)

‘John must be sick.’

(6) GiovanniGiovanni

saràFUT-be

statobeen.

malato.PERF.

(epistemic,sick

past)

‘John must have been sick.’

None of the examples here above is ‘future’, i.e. referring to an event that follows theutterance time. Rather, as can be seen in the translations, these are all modal sentences

3

with an epistemic interpretation. In all these cases, given evidence that the speaker hasat the time of the utterance, he assumes that, in worlds that satisfy certain normalityconstraints, p is true. The future sentences, in this respect are interchangeable withepistemic modals equivalent to must, as indicated also in the translations. See also theexamples below, in Greek and Italian with the actual equivalents of MUST (in Greekprepi takes a subjunctive na-complement, like all modal verbs; Giannakidou 2009):

(7) a. Ithe

AriadneAriadne

prepimust

nasubj

troieat.non-past3sg

tora.now

(epistemic, now)

b. GiacomoGiacomo

oranow

staràFUT-be

mangiandoeat-gerund

c. GiacomoGiacomo

devemust

starbe

mangiandoeat-gerund

’Giacomo/Ariadne must/will be eating now’

(8) Ithe

AriadneAriadne

prepimust

nasubj

milisetalk.past.3sg

xthes.yesterday

(epistemic, past)

‘Ariadne must have spoken yesterday.’

(9) a. GiovanniGiovanni

saràFUT-be

statobeen.

malato.PERF.

(epistemic,sick

past)

b. GiovanniGiovanni

devemust

esserebe

statobee,

malatosick

‘John must have been sick.’

Given this epistemic non-predictive usage, the parallel with MUST, and the lack of futurereading, it becomes very appealing to argue that the future morpheme behaves like anepistemic modal, and this is the line we pursue here, following our earlier works. 1

The future interpretation, which we call predictive, arises with the use of a perfectivenon-past in Greek (10).

(10) a. GianniJohn

arriveràarrive-fut

domanitomorrow

‘John will arrive tomorrow’b. O

TheJanisJohn

thaFUT

ftasiarrive.nonpast.perf.3sg

avriotomorrow

John will arrive at 5pm/tomorrow

We will derive this interoperation in section 4 by a mix of epistemic and metaphysicalanalysis. For now, we focus on the purely epistemic reading.

2.1 Epistemic weakening with future and must : nonveridicality

Giannakidou 2002, to appear argues that FUT in the epistemic non-predictive use func-tions as a weakener of the speaker’s epistemic commitment to the truth of the prejacent.Epistemic weakening is defined as follows (see also Giannakidou to appear):

(11) Epistemic weakening Epistemic weakening is the creation of a nonveridical epis-temic space.

1In previous works we argued that the epistemic interpretation holds with both statives and eventives(and more precisely with perfective non-past in Greek), an issue that we do not discuss here.

4

(12) a. Veridical and nonveridical modal space (Giannakidou 1998, 1999)(i) A modal space (a set of worlds) W relevant to an epistemic agent i isveridical with respect to a proposition p just in case all worlds in W arep-worlds. (Homogeneity).

b. If there is at least one world in W that is a ¬p world, W is nonveridical.(Non-homogenous space).

The spaces we are talking about represent speaker’s knowledge. The idea is thatsentences are not true of false in isolation but truth is always relevant to the knowledge ofan epistemic agent i, also called the individual anchor (Giannakidou 1998, Farkas 1992).We can think of this as a perspectival view on truth, see also Harris and Potts 2009; fora recent discussion of the relevance of individual anchors to temporal interpretation seeMerchant 2013. In an unembedded sentence, the crucial epistemic agent is the speaker.

A veridical modal space is homogenous, as we see, all worlds are p worlds, and thereis no further ordering source or partition to allow doubt (not p worlds). The idea ofnon-homogeneous nonveridical modal space has also been expressed in terms of diversity(Condoravdi 2002). Unlike metaphysical (un)settledness, on the other hand, nonveridi-cality is defined on epistemic alternatives given the individual anchor, and the modal basedoes not include simply facts, but knowledge of the facts attributed to the anchor. This istypically the case with unembedded assertions such as Ariadne is sick. If we compare thissentence to the FUT version of it, we conclude that the speaker, though highly confidentabout the truth of p, in choosing FUT, is still not confident enough to exclude doubt aswitnessed by the test below (from Giannakidou and Mari 2013):

(13) a. I Ariadne ine arosti, #ala dhen ime ke endelos sigouri.‘Ariadne is sick, #but I am not entirely sure.’

b. I Ariadne tha ine arosti, ala dhen ime ke endelos sigouri.‘Ariadne will/must be sick, but I am not entirely sure.’

FUT and modalization in general create nonveridical modal spaces, which are not ho-mogeneous: they contain p and non-p worlds (Giannakidou 1998, 1999). Such spaces ex-press weaker epistemic commitment (see also Smirnova’s 2012 analysis of the subjunctivemood in Bulgarian supporting this point; also Giannakidou, this volume). Weaker com-mitment explains why all modalized sentences, even with a necessity modal, are ‘weaker’than a non-modalized assertion (pace von Fintel and Gilles 2010 who claim that MUSTis strong): there are worlds in the modal base where p is not true, and the actual worldmay turn out to be one of those worlds. 2

Just to make sure that we appreciate this point about the non veridicality ofmust,consider the following case. Context: Ariadne is sneezing, has a fever, watery eyes, etc.

(14) a. Prepimust

nasubj

exihave.3sg

gripi.flu

2A useful recent discussion supporting the nonveridicality of must is found in Lassiter 2013. Lassiteroffers examples like the following:(i) Probably this must have been done before, but I couldn’t find enough information on this in the ISS

docs.(ii) If the handgun was engraved or had some sort of fancier finish then I figured he must be a pistolero.

I might have been wrong but those were my initial impressions.The strength of speaker’s commitment with must here is governed by I figured, so these are clearly

quite weak. There are plenty of similar examples in Lassiter 2013 confirming our view here that must isnot veridical.

5

b. Thamust

exihave.3sg

gripi.flu

c. ThaFUT

prepimust

nasubj.

exihave.3sg

gripi.flu

‘She must have the flu.’d. Exi

She hasgripi.the flu

The doctor is assessing, and given what he knows (the symptoms, his knowledge ofwhat the symptoms mean, the time of the year, etc) he concludes, in c, that Ariadne hasthe flu. In all the worlds compatible with his knowledge/evidence, etc., this is his verdict,he is absolutely certain about it. However, if he choses one of the modal versions, evenwith s must which can combine with tha (as we see in the c example), his modal baseallows also for worlds in which Ariadne does not have the flu (non-p) but e.g. an allergy,or pneumonia. The doctor’s judgment may be that these worlds are not the correct basisfor forming his current diagnose, they are not best (in the sense of Portner 2009); butsuch non-p worlds are indeed in the modal base. The more of those non-p worlds thedoctor allows the less certain he becomes. If the doctor wants to exclude non-p worlds,he must make the stronger statement without must/tha/prepi, that relies on a veridicalepistemic space which is homegeneous: all worlds compatible with what he considers asthe basis for his diagnose are p worlds.

Now, notice crucially that if the doctor is less confident about the diagnose, he willprefer to use the subjunctive particle na with the adverb isos ‘perhaps/possibly’. In thiscase tha is impossible.

(15) IsosPerhaps

na/*thaSUBJ/FUT

exishave.2sg

gripi.flu

‘Maybe you have the flu.’

Note here the translation with ‘maybe’ rather than ‘must’ which is what we used for tha.In choosing to utter the sentence above rather than the tha version, the speaker wants toexpress a weaker judgment, akin to using a possibility modal; Tha, as we see, is impossiblewith isos ‘maybe’, suggesting a ‘threshold’ for tha (a point to which we return).

Tha typically co-occurs with high probability adverbs e.g. malon ‘probably/mostlikely’, and necessity sigoura ‘certainly’, oposdhipote ‘definitely’:

(16) Ithe

AriadneAriadne

malon/profanos/sigouraprobably/obviously/certainly

thaFUT

inebe.3sg

jatros.doctor

‘Ariadne must probably/obviously/certainly be a doctor.’

(17) Ithe

AriadneAriadne

isos/pithanonmaybe/possibly

*thaFUT

inebe.3sg

jatros.doctor

‘Maybe Ariadne is a doctor.’

Na is impossible with these high probability and necessity modals:

(18) Ithe

AriadneAriadne

malon/profanos/sigouraprobably/obviously/certainly

*naSUBJ

inebe.3sg

jatros.doctor

‘Ariadne must probably/obviously/certainly be a doctor.’

6

These contrasts suggest that the force of the modality of tha is stronger than mere possi-bility. As far as Greek in concerned, the same pattern characterizes the necessity modalprepi as we see in the next section. So, Greek appears to have two strategies for epis-temic weakening: one where the speaker has more confidence in his judgment (with tha)and one with less confidence (subjunctive). In both cases particles are used with adverbcombinations. For more discussion of the Greek subjunctive and epistemic weakening, seeGiannakidou, this volume.

In epistemic weakening, the future particle, as well as must/will, look a lot like thewell known German discourse particles such as e.g. ja, wohl (Zimmermann 2011):

(19) MaxMax

istis

(ja)prt

wohlat

aufsea

See. (example from Zimmermann 2011)

‘John must be in the sea.’(For all I know) John will/must be in the sea.

Such particles can also appear in combinations ( ja wohl). Zimmermann makes a claimvery similar to the one we are making here: he says that with wohl, the epistemic com-mitment of the speaker is weakened. Here is a scenario that brings this about:

(20) A: I can’t see Hein.Er ist wohl auf See.‘He must/may be at sea.(21) A: I know for sure:

#Hein ist wohl auf See.Hein is prt at sea

These are exactly parallel to the na and tha versions in Greek we just observed. If I knowfor sure, i.e. if there are no non-p worlds in my epistemic space, I can’t utter Tha inesti thalasa. Giannakidou and Mari make a similar observation about direct evidence: e.g.while raining, I cannot utter FUT p:

(22) Context: I am watching the rain through the window. I say:#Tha vrexi! Greek#Pioverà. Italian‘#It must be raining.’ ‘#It will be raining.’

If we are directly evidencing the rain, then we know that the sentence ‘It is raining’is true; the modalized statement is therefore an infelicitous choice, violating Griceanquantity by saying something weaker than what you know. In Giannakidou and Mari,we took this case to illustrate an evidential component, but generally we can take thisto show incompatibility between the epistemic weakening function of FUT, wohl, will,must, dovere– which require that the speaker be in a nonveridical epistemic state that iscompatible with both p and non-p– and the fact that in direct perception contexts thespeaker in fact knows p. So, we can think of epistemic weakening as a presupposition thatconstrains the distribution of the expressions that have it.

Dutch zullen seems to also function as an epistemic weakener, as we see in the sentencesbelow. The Dutch counterpart of wohl "wel", can also combine with zullen (Giannakidouto appear):

7

(23) A:Hein

Izal

can’t(wel)

seein

Hein.de/op see zijn.

‘He may be at sea (swimming/on a boat).’

(24) A: I know for sure:#Hein zal in de/op see zijn.‘He may be at sea.’

(Many thanks to Jack Hoeksema for discussion of these examples.). We see again acombination zullen wel, just like in Greek and German. We can think of these then ascases of modal ‘concord’ (prima facie).

The Italian facts are parallel, unsurprising. When the speaker knows for sure (i.e. itmodal space is veridical) modal statements are infelicitous.

(25) A: I know for sure:#Deve essere al mare.#Sarà al mare‘He may be at sea.’

However, and importantly, both FUT and the epistemic modal dovere are compatiblewith weak epistemic adverbs, such as forse (see Bertinetto, 1979; Mari, 2010,to appear).Scenario: Giacomo is absent. The teacher tries to figure out why he is absent. She hasvarious options:

(26) a. Sarà forse malatobe-fut maybe sick

b. Devemust

forsemaybe

esserebe

malatosick

c. DovràMust

forsemaybe

esserebe

malatosick

‘(As far as I know), Giacomo must be sick’

The speaker has the choice of using the future (example a) , the epistemic modal (ex-ample b) and a combination of the epistemic modal (c). Each can be combined withthe possibility adverb forse. This shows that across languages modal expressions of thesame category (MUST) support different degrees of non-veridicality: Greek tha/prepi arecompatible only with necessity adverbs, whereas in Italian the range of epistemic adverbscompatible with the future/dovere is wider and includes weaker, possibility adverbs. Ital-ian lacks the weaker commitment subjunctive version of weakening observed in Greek, andfuture/MUST cover the distributions of both tha and na. We argue in the next sectionthat this leads us to refine the notion of ‘modal concord’ as a purely matching strategy.

In Giannakidou and Mari 2012, 2013 we take pains to show that both epistemic weak-ening and the predictive modality of FUT rely on knowledge of the speaker, so they areprofoundly epistemic in this sense. But the purely epistemic reading does’t make a pre-diction about the future. The important generalization about epistemic weakening hereis that future words crosslingusitically, regardless of their status as bound morphemes(Italian), particles (Greek), or auxiliaries (English, Dutch) all seem to have purely epis-temic uses without reference to the future. And in their function as epistemic weakenersthe futures are similar to modal particles of Dutch and German, as well as must and itsequivalents in Greek, Italian, and Dutch.

8

We now turn to considering the role of adverbs, and to our new proposal about ‘modalmodulation’.

2.2 The force of FUT, and the role of adverbs: modal modulation

Now, recall some core examples in the epistemic use.

(27) a. IThe

AriadneAriadne

thaFUT

kimatesleep.INP.3sg

(tora).now

(Greek)

‘Ariadne must be sleeping now.’b. I

TheAriadneAriadne

thaFUT

inebe.INP.3sg

giatros.doctor

‘Ariadne must be a doctor.’c. I

theAriadneAriadne

thaFUT

peziplay.INP.3sg

(tora).now

‘Ariadne must be playing now.’

(28) a. Giacomo(Italian)

dormiràGiacomo

(adesso).FUT-sleep (now)

‘Giacomo must be sleeping (now).’b. Giacomo

GiacomosaràFUT-be

dottore.doctor.

‘Giacomo must be a doctor.’c. Giacomo

GiacomogiocheràFUT-play

(adesso).(now).

‘Giacomo must be playing (now).’

In all cases, the reading is inferential: I am considering information I have, and drawan inference based on that information. For example, with regard to the a examples, Iknow that Ariadne/Giacomo have the habit of taking a nap at 2pm, I also know thatthey always stick to schedule, and I also know that today has been a regular day. So at2:10 pm, I utter the sentence expressing my relative certainty that Ariadne and Giacomoare, for all I know, asleep. Likewise, I know also that usually by 6 pm Ariadne is downat the yard playing with her friends. At any time after 6 pm then, I can utter (27-c) withthe same degree of certainty. Finally, (27-b) expresses an inference based on evidence: Ihave witnessed Ariadne expressing opinion on medical matters, she cites reliable medicalsources all the time etc., hence I am entitled to conclude (27-b).

In the examples above, notice the order FUT > non-past, especially visible in Greek,where the FUT is an independent particle preceding the tensed verb. We will assumethat this order reflects the actual scope order. When scoping over PAST, FUT also hasa similar epistemic interpretation, only now we are assessing a past event:

(29) a. Ithe

AriadneAriadne

thaFUT

kimithikesleep.PP.3sg

(prinbefore

apotwo

dyohours

ores).

‘Ariadne must have fallen asleep two hours ago.’b. I

theAriadneAriadne

thaFUT

milisetalk.PP.3sg

(xthes).yesterday

‘Ariadne must have talked yesterday.’

(30) Giacomo si sarà addormentato due ore fà.Giacomo reflexive FUT-be fallen-asleep two hours ago.

9

‘Giacomo must have been fallen asleep two hours ago.’

Again, knowing Giacomo’s or Ariadne’s habits allows me to infer now that at somepoint two hours ago, they fell asleep. The past adverbial ‘two hours ago’ functions as thetime frame for the eventuality, taking scope below FUT.

Let us now turn to the combinations between the particle and the adverbs. Thesecombinations not only further plea for considering the future a modal rather than atemporal operator, but also allow us to propose a new way of disentangling differentcontributions of epistemic expressions. The term ‘modal concord’, for the phenomena weare studying, is a misnomer because the relation between the adverb and FUT is not oneof identity or even matching, as is understood from work on modal concord such as e.g.Huitink 2012.

As we already noted in the previous section, tha co-occurs with high probability ad-verbs e.g. malon ‘probably/most likely’, and necessity sigoura ‘certainly’, oposdhipote‘definitely’ - but is bad with mere possibility adverbs such as isos ‘maybe/perhaps’ andpithanon ‘possibly’:

(31) Ithe

AriadneAriadne

malon/profanos/sigouraprobably/obviously/certainly

thaFUT

inebe.3sg

jatros.doctor

‘Ariadne must probably/obviously/certainly be a doctor.’

(32) Ithe

AriadneAriadne

isos/pithanonmaybe/possibly

*thaFUT

inebe.3sg

jatros.doctor

‘Maybe Ariadne is a doctor.’

With possibility adverbs, the weakening is done with the subjunctive, as we said. Theincompatibility with possibility adverbs suggests that the force of the modality of tha isstronger than mere possibility. As far as Greek in concerned, the same pattern character-izes the necessity modal prepi (the only necessity modal verb in Greek), which combineswith the same range of adverbs, but also with FUT itself:

(33) IAriadne

Ariadneprobably/#possibly

malon/#isosFUT

thamust

prepisubj

naleft.PNP.3sg

efije.

‘Ariadne ?probably/#possibly must have left.’

(34) TaThe

pediachildren

prepimust

oposdhipotedefinitely

naSUBJ

inebe.

sto3pl

spiti.in-the home

Epistemic necessity: ‘The children must definitely be at home.’

(All complementation is finite in Greek, and modal verbs take subjunctive na comple-ments). Prepi is compatible with adverbs above a certain threshold of high probabilityreaching to necessity - so indeed, in this case we can talk of matching modalities. Prepiis not so different from English must as is shown in the examples in Lassiter 2013; seealso Giannakidou 2012 for more examples such as You must probably think I am crazy.This more flexible behavior of Greek and English MUST, along with Italian dovere (whichappears to be even more flexible, as we saw) challenges the necessity of the Kratzerian dis-tinction between should (weak necessity) and must (strong necessity). Natural languagesMUSTs seem to be more flexible and typically convey both.

It would be helpful to see how these data fit with theories of gradable modality suchas e.g. Lassiter 2011. Lassiter suggests that modal sentences get their truth valuesby comparing the position of their proposition on a relevant scale to a threshold value

10

(determined by context and the lexical semantics of the modal). In this assumption,MUST works as follows:

(35) [φ must be the case ]M,w,g = 1 if and only if must (φ) is greater or equal to θmust

(Lassiter 2011: (1.6))

We have here a measure for what counts as MUST - the threshold value θmust. Thisthreshold may differ from one language to another - recall that in English must is harder(though not impossible) with probably, but in Greek, combinations of malon with prepiand tha are common. So, the threshold for MUST in Greek is lower and includes weaknecessity (or high probability). Modal concord, in this frame, can be restated as specifyingsubspaces on the measurement scale, a guide of where to look for the numerical value oftha p: e.g. on a scale from 0 to 1, with malon, we must look at values of high probability,say between .8-.9; with oposdhipote, we only look at .99. The adverbs therefore serve asprobability restrictors (as argued in Giannakidou 2012). Our current view of the role ofadverbs sections 3 and 4 will preserve this idea, but we will claim further that the adverbsdo not compose syntactically with the modality but only have presuppositional content.3

As we said earlier, the Italian facts are parallel, yet not identical to Greek in thatthe Italian FUT is not categorically excluded with weak possibility modals like forse(Bertinetto, 1979; Mari, 2009,2010, to appear):

(36) ForseMaybe

GiacomoGiacomo

saràFUT-be

una

dottore.doctor.

‘Giacomo may possibly be a doctor.’

This sentence expresses a degree of uncertainty in the assessment due to the weak adverb.The statement appears to be closer to the na-versions we discussed for Greek. So, wemay want to say that though Greek has a threshold value for FUT, Italian doesn’t. Whenpredicting, even Greek removes the threshold (section 4).

Italian FUT is also good with stronger modals, just like in Greek, (Bertinetto, 1979,Mari 2010).

(37) Probabilmente Giacomo sarà un dottore.Probably Giacomo FUT-be a doctor‘Giacomo must probably be a doctor.’

But in Italian, unlike Greek, the future morpheme and the epistemic modal dovere areindeed compatible with weak epistemic adverbs, such as forse. So, again, the facts for thefuture and the epistemic verb are parallel in Italian, as both (as well as their combination)are compatible with weak epistemic adverbs. Scenario: Gicaomo is absent. The teachertries to figure out why he is absent. She has various options:

(38) a. Sarà forse malato3The adverbs could also be seen as degree modifiers, as entertained in Giannakidou 2012, following

a suggestion by Dan Lassiter. If modal adverbs are probability degree modifiers, the differences inacceptability as we move down the range of epistemic modals (ordered by strength) become similar to thecontrast between slightly damp/wet/#soaking or somewhat large/#enormous or totally #ajar/open. Themodal-adverbial relation becomes thus parallel to the degree modifier-adjective relation in establishingpermissible and non-permissible combinations that rely on the combinatorial restrictions of the modifiersas they ‘match’ (or do not match) the probability space established by the modal operator. This is veryclose to the view of the role of adverbs we have here.

11

be-tut maybe sickb. Deve

mustforsemaybe

esserebe

malatosick

c. DovràMust

forsemaybe

esserebe

malatosick

‘(As far as I know), Giacomo must be sick’

In Italian, FUT can also combine with dovere. This combination is in turn compatiblewith weak adverbials such as forse. Consider a context in which I suspect that my sonhas heard a secret conversation between me and my husband. I can utter:

(39) Dovràmust-FUT

forsehave

averheard

sentitoeverything

tutto.

He must have heard everything’

Indeed, the combinations FUT dovere are weaker in Italian, with an even higher flexibilitythan in Greek.

To summarize: we saw that Greek and Italian FUT have epistemic non-predictivereadings. Typically, they associate with high confidence about the prejacent proposition,but in Italian they are more flexible and this flexibility leads to modal modulation ratherthan modal concord. Languages vary with respect to whether they place a threshold:Greek does, but Italian either has a lower threshold or no threshold at all, since necessitymodals and FUT are supported by weak possibility adverbs. In our analysis in section4, we make the role of adverbs and their relation to the modality more precise. Here wejust want to emphasize that their active appearance with future morphemes, and the richcombinations we observe, clearly argue in favor of epistemic modality.

3 Predictive reading and gradable strength of predic-tion

Reference to a future event, which we call here the ‘predictive reading’ is done with theparticular combination of FUT plus a lower non-past (which in Greek must be perfective).

(40) OThe

JanisJohn

thaFUT

ftasiarrive.PNP.3sg

stisat

55pm/avrio.pm/tomorrow.

(predictive reading)

’John will/#must arrive at 5 pm/tomorrow.’

(41) GianniGianni

arriveràFUT-arrive

alleat

55delpm.

pomeriggio. (predictive reading)

Gianni will/#must arrive at 5 pm.

As we see, the paraphrase with must is pretty odd in this case.Because of the adverb, the speaker has specific temporal information about where to

place the eventuality. The future reading is forward shifting of the eventuality time dueto the adverb scoping above the TP but lower than the FUT:

(42) FUT > alle 5 del pomeriggio > arrive

Though non-veridical - at the time of knowledge/assessment, the event hasn’t happenedyet - the forward shifted, predictive reading of FUT obtained with a future adverb, does

12

seem to express some certainty that the event will happen. This might lead one to assumethat the predictive reading is always strong, that it expresses always very high confidenceabout the future course of events. However, just like in the purely epistemic reading, FUT(including will) is compatible with adverbs of various strength, including some really weakones:

(43) Arriverà forse/probabilmente/certamente alle 4.‘He will maybe/probably/certainly arrive at 4’

(44) O Janis tha erthi sigoura/malon/isos stis 5.‘John will definitely/probably/maybe arrive at 4’John will definitely come at 4.John will probably come at 4.Maybe John will come at 4.

(45) Gianni arriverà sicuramente/probabilmente/forse alle 4.‘John will definitely/probably/maybe arrive at 4’

Recall from our earlier discussion that weak, existential modal adverbs such as forse arecompatible with the non-predictive use in Italian but not in Greek. In the prediction,however, we notice more flexibility in Greek as well: isos ’maybe’ is now OK. Will showsthe same pattern, and can combine, just like the Greek and Italian futures, with some-thing as weak at maybe. Things get even more ’loose’ when we consider low-probabilityadverbials (like 10% chance, 30% chance):

(46) Kata 10%, 30%, 90% O Janis tha erthi stis 5.by 10% 30% the John FUT come at 5There is 10% probability that John will come at 5.There is 30% probability that John will come at 5.There is 90% probability, that John will come at 5.

(47) Ci sono 10%, 30%, 90% di chances che Gianni arrivi alle 5There is 10% probability that John will arrive at 5.There is 30% probability that John will arrive at 5.There is 90% probability, that John will arrive at 5.

So, when we predict, we seem to be able to use probability measures of weak force toindicate the degree of confidence we have in our prediction. This is an important point tomake, and to our knowledge it has not been noticed before. The indication of probabilityis highly sensitive to what the speaker knows or takes as a reliable information, and issuggestive of an epistemic layer, we will argue, even in the prediction.

What will happen in the future is objectively unsettled at the present time (Kaufmann2005, Kaufmann et al. 2006) and therefore objectively non-veridical - since the futureevent hasn’t happened yet. However, the future is knowable in this way: speakers makeinformed predictions about the likelihood of propositions being true at a future time,based on what they know at the present time. The flexibility observed with adverbsallows us to say that the predictions fall generally into two categories: predictions ofhigh confidence (based on adequate knowledge that allows such high confidence), andpredictions based on low confidence (based on less adequate knowledge). One can thinkof such low confidence predictions as ’guesses’. Speakers can of course also have averageconfidence. In our semantics next, we will try to capture these ideas in a precise way, byusing the notion of reasonable future (Mari, 2013). Once we define that, we will claim

13

that the adverbs function as in the epistemic weakening case, measuring the speaker’sconfidence about the actual world to come being in the set of reasonable futures.

4 Analysis: modal adverbs as confidence modulators inthe epistemic and the predictive reading

While we acknowledge the irreducible asymmetry between a fixed (settled) past and anon-settled futures, our semantics captures the role of adverbs as modulators uniformlyin the epistemic and the predictive reading.

Let us start with the epistemic case, in which p is evaluated either in the present orin the past.

4.1 Analysis for the purely epistemic reading of the future

The background of our analysis is Giannakidou 2009, and Giannakidou and Mari, 2012a,b.An important premise that we adopt from these works is that the lower tense, and not FUTitself contributes the temporal information (past or nonpast). We saw in our earlier datain both Greek and Italian that the lower tense is responsible for locating the eventualityin the past or non-past.4

Giannakidou says that Greek perfective nonpast denotes an interval whose left bound-ary is a dependent (Giannakidou 1998) variable t (for a given time t, [t,∞) stands for theinterval that starts at t and goes on to infinity).

(48) [[perfective non-past]] = λpλt(P ( [t,∞ ))

This meaning of the nonpast is close to Broekhuis and Verkuyl’s 2013 analysis ofextended present. In their theory, the conceptual ‘present’ is a an extended interval thatincludes now but stretches after it. With the non-past (as opposed to a true present), theleft boundary needs to be negotiated as now, and this requires additional structure, asGiannakidou argues. This structure is given by nonveridical particles in Greek, includingtha, the optative as and the subjunctive na– all with prospective, future orientation(see Giannakidou 2009, this volume for discussion). In this analysis then, the particles,including tha, bring in the speech time, and this is a premise we adopt. The variable t in(48) can thus be set as being tu when tha combines with the tensed proposition.

(49) [[tha]] = tu

One could also say that the additional particle structure triggers an "Insert PRES"rule– since every sentence is evaluated temporally wrt the speech time. So, the presentin the particle nodes may be some kind of default contribution, common to all particles.Tha, semantically, is a modal, and indeed a universal one:

(50) For any world w, conversational backgrounds f, g and epistemic agent i:[FUT]w,f,g,i = λq < st > .∀w′ ∈ Bestg(w)(∩f(w)) : q(w′) = 1where Bestg(w)(X) selects the most ideal worlds from X, given the ordering g(w)determined by i

4For English, Condoravdi 2002 claims that modals are both temporal/modal and the lower tense doesnot contribute anything.

14

In previous work, we have shown that this analysis applies to Italian as well (Giannakidouand Mari, 2012a,b).

We assume the theory of modality in Kratzer (1981, 1991), Portner (2009), and Hac-quard (2011), with two conversational backgrounds as arguments of a modal expression -the modal base and the ordering source. The modal base f is the factual background, andthe ordering source g is a normative background. With FUT, in this non-predictive use,the modal base is epistemic; specifically, it is the set of propositions known by the agenti, i.e. the speaker in an unembedded context: ∩fepistemic(w) = λw′.w′ is compatible withwhat is known by i (the speaker) in w.

The ordering source g(w), on the other hand, orders the worlds in ∩f(w) accordingto normalcy assumptions, following Kratzer and Portner. Modal expressions of necessityquantify over those worlds that adhere to the norms in the ordering source as much aspossible. We can call these worlds Best following (Portner, 2009). What we mean bynormalcy conditions, is that strange things do not happen (see Portner, 1998 and discussionin section 4). For instance, if I have red cheeks and sneezing nose, then, under normalcircumstances, I have the flu. However, circumstances are not necessarily normal. In suchextraordinary circumstances these symptoms are secondary and indeed indicative of apotentially much worse disease. Typically the actual world tends to be non-extraordinary,but we also know that strange things happen. As a consequence, we do not claim that theepistemic agent actually knows that the actual world belongs to the set of best worlds.Given, on the other hand, that the accessibility relation is epistemic and therefore reflexive,it is ensured that the actual world is in the modal base. What it is not ensured, anddepends on the ordering, is that the actual world is also in the set of best worlds.

Only in the Best worlds is p true, therefore the modal (universal or existential) isnonveridical: the modal base is a non-homogenous space containing p and ¬p worlds. So,a modalized sentence, even with a necessity modal like FUT or must is ‘weaker’ than anon-modalized assertion: there are worlds in the modal base ∩f(w) where p is not true,as we see, and we are not sure where the actual world is, i.e. in the best worlds or not.

Now, putting the temporal component of our analysis (that FUT makes reference totu) together with the modal gives us the following truth conditions, for the scopings FUT> nonpast and FUT > past respectively; (∩f(w, tu)) provides the alternatives availableto the speaker at the utterance time, and q(w′, t′) states that q is true in the projectedalternatives at time t′ that either coincides or follows the evaluation time tu:

(51) [FUT non-past/tha PNP]w,f,g,i = 1 iff: ∀w′ ⊂ Bestg(w)(∩f(w, tu)) : ∃t′ ∈[tu,∞) ∧ q(w′, t′) = 1where Bestg(w)(X) selects the worlds that are ranked as best from X, given theordering g(w).

(52) [FUT past/tha PAST]w,f,g = 1 iff: ∀w′Bestg(w)(∩f(w, tu))) : ∃t′ ≺ tu∧q(w′, t′) =1where Bestg(w)(X) selects the worlds that are ranked as best from X, given theordering g(w).

As we made clear earlier, this does not imply anything about the actual world belongingto the best worlds. This is where the adverbs come into play.

The first thing we will say is that the adverbs do not directly compose with p orFUTp. The reason we say this is rather practical: given the variability in force we saw,it is impossible to combine FUT (a universal) with the adverbs in a concord like manner.Even the most flexible theories of modal concord (such as Huitink 2012) would require

15

at least matching force. So, the adverb is forced to contribute at the non-at issue level,something consistent with the fact that modal adverbs are speaker oriented (Ernst 2009)and seem to ‘scope’ high in the sentence (Ernst therefore argued that they are positivepolarity items).

We will claim that the modal adverbs contribute a presupposition that there is ameasure of confidence in the assessment of the epistemic agent i which measures howconfident i is that the actual world will be within the set of the best worlds. One couldprobably see this as a generalized contribution of epistemic modals - and perhaps evenmodals in general: they do not simply make a claim about the possible worlds, but bearrelevance to whether an agent i thinks that the actual world is one that conforms to thebest ordering, and to what degree. So here is how we put this all together:

Non-predictive use of FUT (= future for past and present)

(53) FUT p asserts necessarily p, relative to an epistemic modal base and an epistemicagent i.Presupposition: there is a measure function µconfidence determined by i that mea-sures how confident i is that the actual world will be within the set of the bestworlds. The default value of µconfidence will range above 70%.Truth condition: p is true only in the best worlds, given the evidence: ∀w′ ∈Bestg(w)(∩f(w)) : p(w′) = 1.

As indicated in the presupposition, we will assume that bare FUT, with no adverb, con-tributes a default of high confidence. But with the use of adverbs, the value becomesmore specific and can express higher or lower confidence. This captures the variability inthe epistemic judgement between very high, middle, and low confidence.

Finally, and in support of this analysis, it is important to note that the confidencepresupposition survives negation.

(54) a. GianniGianni

nonnot

arriveràFUT-arrive

micaeven

‘Gianni must not be arriving.’

b. GianniGianni

nonnot

saràFUT-be

malatosick

‘Gianni will not be sick / John must not be sick’

The speaker retains the default degree of high confidence that the actual world is not inthe set of best worlds that make p true. And is also consistent with the observation thatepistemic modals generally scope above negation (e.g. Cinque, 1999; Hacquard, 2006;Homer, 2010) and likewise modal-evidentials across languages, e.g., Japanese (McCreadyand Ogata 2007).

4.2 The predictive reading

4.2.1 The general picture

We have claimed that the future is an epistemic in both the non-predictive and thepredictive use. However, in order to keep a parallel analysis for the non-predictive andthe predictive readings, one would have to assume that on the latter reading also themodal base contains worlds in which some facts are true and these same facts are known

16

to be true by the speaker. However, since the future has not occurred yet, the modal basecannot be epistemic, in the sense that alternatives cannot be known by the speaker.

What is thus the role that knowledge plays in the predictive use?The speaker, given what he knows at the utterance time, considers only a subset of

available metaphysical future alternatives. Given the specific normality constraint thatwe will design (we will elaborate on the notion of normality here below), it turns out thatthe speaker knows, at the time of the utterance that, in these futures, p will be true. Whatit is inexorably not known is whether the actual-world-to-come will be in this restrictedset of alternatives. We thus claim that a prediction is an assessment of the speaker’sconfidence that the actual-world-to-come will be one of the alternatives in which p comesout true.

Put it in other terms, we propose that, making a prediction means: (i) determiningvia present knowledge what are the metaphysical alternatives in which p comes out true;(ii) not knowing whether the actual world to come will be one of these metaphysicalalternatives and (iii) believing or being convinced that the actual world to come will beone of these metaphysical alternatives.

This view contrasts with a possible position holding that a prediction involves quan-tification over doxastic alternatives (Mari, 2009,to appear), i.e. a view according to whichthe speaker has a present belief or knowledge about what the actual world will be. Suchview would have two major shortcomings, one theoretical and a second one empirical.

At the theoretical level, one cannot hold that on the predictive reading the interpre-tation is epistemic in the same way as it is with the non-predictive reading. As we haveexplained, in the latter case facts and knowledge of facts is used. In the predictive readingneither facts nor knowledge of facts is available, as facts have not occurred yet and thuscannot be known (see Kaufman, 2005, Mari, 2013). Hence, the ‘epistemic/doxastic’ viewwould boil down to consideration of beliefs or preferences of the speaker about what theactual-world-to-come will be like, and that would make a very weak theory of prediction–as it would lose the connection between knowledge and facts, a connection that we wantto maintain. At the empirical level, a doxastic view would thus be too weak and toostrong at the same time: we would allow beliefs to determine outcomes (too weak), whilebeing unable to explain the varying distributions of the adverbs (too strong).

4.2.2 Reasonable futures and the actual-world-to-come

What are the alternatives in which p is true in the future ? As the future is not settledcan these be known ? Our answer is yes, and these are the reasonable futures (Mari,2013).

What is a reasonable future ? Following Mari (ibid.), a reasonable future determinedat time t is such that the rules that hold at t are maintained. As FUT sets the timeof the perspective at present, the time at which reasonable futures are considered is thetime of the utterance. Rules can be social (determined by convention), natural (the lawsof nature), or moral (laws that are conform to the commonly admitted accepted moralbehavior). They also include habits (i.e. behavior patterns). Rules, especially social ones,can be restricted to very specific contexts as well (e.g., rules that hold in my office, myhome, and my building). Consider (55).

(55) Gianni arriverà a Roma con la macchina alle 4pmJohn will arrive in Rome by car at 4pm

17

The rule that matters when interpreting the sentence can be summarized as, ‘if you usea car, you reach the intended destination in a certain amount of time’. When evaluatingthis sentence, one disregards those futures in which accidents, breaking of the car andsimilar aberrant events occur (see also Dowty’s 1979 notion of inertia), which wouldcause the car not to reach the destination.

Note that, because social, moral and even natural aberrations occur very often (seediscussion in Landman, 1992), the world that will become the actual world could neverbelong to the set of reasonable futures of any time. When evaluating sentence (55), thoseaberrations that would affect the individuals in the reference set are not considered (e.g.,when interpreting (55), one does not consider an earthquake occurring in Asia, as this,presumably, would not affect a travel to Rome).

Moreover, when interpreting (55), one also disregards futures in which aberrations(such as attacks and earthquakes) do not occur but in which the rules that hold at thetime at which reasonable futures are considered are nonetheless modified. For example,when evaluating (55) at time t0, one disregards those futures in which the driver decidesto stop in a different city and remains there for three days.

Reasonable futures are such that also habits and behaviors do not change there. Ofcourse habits can change. However, those futures in which they change are discarded (seealso discussion in Portner, 19885).

In our account, we consider time-world coordinates and use sets of times and worlds.We assume that, for any given time, the past of a world at that time is unique, whereasthe not-yet-realized-future of that world at that time, is one of the options available att (Thomason, 1984; Condoravdi, 2002; Kaufman, 2005).6 In particular, considering theutterance time t0, we assume that the actual world up to and including t0 is unique butthat there is not yet an actual world after t0 and that a variety of options are available.For any time t, we call a ‘reasonable world-to-come’ at t, a ‘reasonable future’ fixed att. In our framework, the time t at which reasonable futures are considered can eithercoincide with or precede the utterance time. It is possible that the actual world-to-comeat t belongs to the set of reasonable futures fixed at time t.

Along the lines of Mari (ibid.), we now elaborate the modal framework that we employ(for an extended discussion, see Mari, ibid.).

We use aW×T forward-branching structure (Thomason, 1984). A three-place relation' on T ×W ×W is defined such that (i) for all t ∈ T , 't is an equivalence relation;(ii) for any w,w′ ∈ W and t, t′ ∈ T , if w′ 't′ w and t precedes t′, then w′ 't w (we usethe symbols ≺ and � for temporal precedence and succession, respectively). In words, wand w′ are historical alternatives at least up to t′ and thus differ only, if at all, in what isfuture to t′.

For any given time, a world belongs to an equivalence class comprising worlds withidentical pasts but different futures.

For any time t ∈ T , we define the common ground cg(t) as the set of worlds that areidentical to the actual world w0 at least up to and including t.7

(56) cg(t) := {w | w 't w0}

So defined, the common ground includes any world branching from the actual world at5We do not use Portner’s framework, as unlike the author, we do not use events in our account, and

the notion of reasonableness is relativized to events in Portner’s theory.6We present the formal framework of branching time in section ??.7See also Condoravdi, 2002 .

18

a time equal to or after t, including those worlds that are highly different in their causallaws from the actual world as well as those worlds in which current causal and social lawsmalfunction.

In this framework, Mari (2013) defines ‘reasonable futures’ as in (57). For any t ∈ T ,

(57) ReasFut(t) := {wi ∈ cg(t) | wi is such that the set of rules fixed at tcontinue to hold inwi}

The set of reasonable worlds determined at a certain time t is different according todifferent epistemic agents, and the facts that they take into account at the time of theuuterance. Hence, amending (57), we define (58), where epistemic agents are added asparameters.

(58) Given the epistemic agent i, ReasFut(t) :={wi ∈ cg(t) | wi is such that the set of rules fixed at t and considered by icontinue to hold inwi}

Let us note here, the actual world w0 is not an already existing complete history (seeMari, 2013 for a more detailed discussion). Let us consider the utterance time tu. Attu, one can state what the reasonable futures of time tu are, however, one cannot statewhether the actual-world-to-come belongs to the set of reasonable possibilities, since theactual-world-to come does not exist yet at tu.

Reasonable futures are not a projection of speakers’ preferences and beliefs; they aremetaphysical alternatives that do not exist yet at the time of the utterance.

Moreover, as we have have explained, one cannot state at the time of the utterancewhether the actual-world to come will be reasonable or not. At the utterance time, weknow whether the actual world is reasonable or not until the utterance time, but we donot know whether the actual-world-to-come will be reasonable after the utterance time.

We claim that, when stating a future sentence, on the predictive reading the speakerchooses to consider, based of its present knowledge, only reasonable futures on the, asubset of the common ground. This does not mean that he knows that the actual world-to-come will be reasonable, as the actual world to come is not there yet, and therefore,cannot be known. However, the speaker has confidence that the actual world to comewill be a reasonable one. In the bare case (John will come at 4pm), the assumption thatthe actual world to come will coincide with a reasonable future at 4pm is ranked veryhighly, which means, associated with a high degree of confidence. This means that FUTitself must carry by default high confidence, consistent with what we observed with thenon-predictive reading. On the other hand, when we have an overt adverb, it is the adverbthat provides the measure of confidence that the actual world to come will be reasonable(i.e. will be in ReasFut(tu)). For instance, if the adverb is probably, and the measureinterval is between 1 and 100, then probably(FUT(p)) says that, according to the speaker,the probability of the actual world to be a reasonable future where p is true is between85 and 100 percent.

We are now ready to provide an analysis for the predictive use of the future, which isin (59):

(59) FUT(NON-PAST(p)) = 1 iff ∀w′ ∈ ReasFut(tu) : ∃t′ ∈ [tu,∞(∧p(w′t′)Presupposition: There is a measure function that measures the speaker’s confi-dence that the actual-world-to-come will be in ReasFut(tu). The default value ofthe adverb is above 70%.

19

Truth condition: p is true only in the the reasonable futures, given the evidence:∀w′ ∈ Bestg(w)(∩f(w)) : p(w′) = 1. All reasonable futures are p futures.

To sum up: in the purely epistemic reading, the future morphemes operate on epis-temic alternatives, and have the expected ordering that comes with epistemic modals. Soin all respects, they behave like epistemic modals. In our predictive analysis, we usedmetaphysical, not epistemic alternatives. The epistemic layer here is detected in the con-tribution of adverbs, which will now measure how confident the epistemic agent is thatthe actual world to come will be a reasonable one. So this epistemic presupposition iswhat unifies both uses of the future.

5 ConclusionsThe major lessons in the present paper are two. First, future particles and verbs, as well asepistemic modals such as MUST are used systematically in a number of languages (Greek,Italian, English, Dutch) as epistemic weakeners, to express weaker epistemic commitmentof the speaker towards the propositional content. As a class then, MUST and futuresin this function are non veridical (pace von Fintel and Gilies 2010, and in support ofKarttunen’s original weak analysis, and Giannakidou’s nonveridicality thesis 1998, 1999).The second lesson is that in both predictive and epistemic uses of the future morphemes,the modal adverbs contribute a presupposition that there is a measure of confidence in theassessment of the individual anchor i which measures how confident i is that the actualworld will be within the set of the best, or more reasonable worlds. This insight, we wouldlike to suggest, probably describes a generalized ingredient of epistemic modals: they donot simply make a claim about the possible worlds, but bear relevance to whether anagent i thinks that the actual world is one that conforms to the best ordering, and towhat degree. In the predictive reading, we argued that both epistemic and metaphysicaldimensions are needed. The epistemic dimension again is revealed by the role of modaladverbs. In both uses, the adverbs come about as modulators of the speaker’s confidence,which seems to be the ubiquitous epistemic ingredient of the future.

References

Bertinetto, P.M. 1979. Alcune Ipotesi sul nostro futuro (con alcune osservazioni supotere e dovere), Rivista di grammatica generativa 4 : 77-138.

Broekhuis, H. and Verkuyl, H. (to appear). Binary Tense and Modality. NaturalLanguage and Linguistic Theory.

Blaszack, J. and Dorota Klimek-Jankowska] 2012. Futures in Polish and Slovenianfrom the perspective of a force-dynamic model. In Aguilar-Guevara, A., A. Chernilovskaya,and R. Nouwen (eds.),Proceedings of Sinn und Bedeutung 16: The Utrecht Meeting, 2011.(MIT Working Papers in Linguistics). Cambridge, MA: MIT Press, 85-98.

Chiou, Michael. 2012. The interpretation of Greek future: a pragmatic approach. Ms.Cinque, G. 1999. Adverbs and Functional Heads. Oxford: OUP.Condoravdi, C. 2002. Temporal interpretation of modals: modals for the present and

for the past. In The Construction of Meaning, eds D. Beaver, Luis D. Cassillas Maritinez,Brady Z. Clark, S. Kaufmann.

20

Copley, B. 2002. The semantics of the future, PhD MIT.Dowty, David 1979. Word Meaning and Montague Grammar. Dordrecht: Kluwer

Academic Publishers.Ernst, Thomas 2009. Speaker oriented adverbs. Natural Language and Linguistic

Theory 27, 497-544.Faller, M. 2002. Semantics and pragmatics of evidentials in Cuzco Quechua. PhD

thesis, Stanford University.Farkas, D. 1992. On the Semantics of Subjunctive Complements, in P. Hirschbühler

(ed.), Romance Languages and Modern Linguistic Theory, John Benjamins, pp. 69-105.von Fintel, K. and Gillies, A. 2010. Must...stay... strong ! Nat. Language Semantics

18 : 351-383.von Fintel, K. and Heim, I. 2007. Intensionality. Ms. MIT.von Fintel, K. and Iatridou, S. 2006. How to Say Ought in Foreign: The Composition

of Weak Necessity Modals. Ms. MIT.Giannakidou, A. 1998. Polarity Sensitivity as (Non)veridical Dependency. John Ben-

jamins. Amsterdam.Giannakidou, A. 1999. Affective dependencies. Linguistics and Philosophy 22 : 367-

421.Giannakidou, A. 2009. The dependency of the subjunctive revisited: temporal seman-

tics and polarity. Lingua 120: 1883-1908.Giannakidou, A. 2012. The Greek future as an epistemic modal. In the Proceedings

of ICGL 10.Giannakidou, A. to appear. The modality of the present and the future: Greek, Dutch,

and beyond. To appear in Natural Language and Linguistic Theory.Giannakidou, this volume. The landscape of subjunctive: epistemic and evaluative

dimensions. Giannakidou, A. & Mari, A. 2012a. Italian and Greek futures as epistemicoperators. to appear in CLS 48.

Giannakidou, A. & Mari, A. 2012b. Italian and Greek futures as evidentials operators.to appear in Proceedings of Sinn und Bedeutung 17.

Giannakidou, A. & Yoon, S. 2011. The subjective mode of comparison: metalingusiticcomparatives in Greek and Korean. Natural Language and Linguistic Theory 29:621-655.

Hacquard, V. 2006. Aspects of Modality. PhD MIT.Hacquard, V. 2011. Modality. In C. Maienborn, K. von Heusinger, and P. Portner

(eds.) Semantics: An International Handbook of Natural Language Meaning. HSK 33.2Berlin: Mouton de Gruyter, pp. 1484-1515.

Harris, Jesse A. and Christopher Potts. 2009. Perspective-shifting with appositivesand expressives. Linguistics and Philosophy 32(6):523-552

Huitink, J. 2012. Modal concord. A case study in Dutch. Journal of Semantics 29(3):403-437.

Homer, V. (2010). Epistemic Modals: High ma non troppo. Proceedings of NELS 40.Kaufmann, S. 2005. Conditional Thruth and Future Reference. Journal of Semantics.

doi: 10.1093/jos/ffh025Kaufmann, S., Condoravdi, C. Valentina Harizanov, V. 2006. Formal approaches to

modality. In W. Frawley (ed.), The Expression of Modality, Mouton de Gruyter., 71-106.Kissine, M. 2008. From predictions to promises. Pragmatics and Cognition 16 :169-

189. KKratzer, Angelika. 1981. The Notional Category of Modality. In Words, Worlds, and

Contexts. New Approaches in Word Semantics, ed. By H. J. Eikmeyer & H. Rieser, 38?

21

74. Berlin: de Gruyter.Kratzer, A. 1991. Modality. In Semantics: An International Handbook of Contempo-

rary Research, ed. By A. von Stechow & D. Wunderlich, 639-650. Berlin: de Gruyter.Lassiter, D. 2011. Measurement and Modality: the Scalar Basis for Modal Semantics.

PhD thesis, NYU.Lassiter, D. 2013. On the weakness of must: in defense of the mantra. Paper presented

at the CSLI workshop Perspectives on Modality. Stanford University.Matthweson, L. 2010. Crosslinguistic variation in modality systems: the role of mood.

Semantics and Pragmatics 3 : 1-74.Mari, A. 2009. Disambiguating the Italian Future. In Proceedings of Generative

Lexicon, 209-216.Mari, A. 2010. On the evidential nature of the Italian future. Ms. IJN http://

jeannicod.ccsd.cnrs.fr/docs/00/67/85/49/PDF/Evidential_Future_Italian.pdfMari, A. 2013. Each other, asymmetry and reasonable futures. Journal of Semantics.

doi: 10.1093/jos/fft003Mari, A. to appear. On the modal meaning of Italian future tense. Lingvisticae

Investigationes.McCready, E. and N. Ogata. 2007. Evidentiality, modality, and probability. Linguis-

tics and Philosophy 30 :147-206.Merchant, Jason. 2013. Individual anchors for tenses; how Keats learned to read

before Shakespeare. Linguistic Analysis 38. Murray, Sarah. 2010. Evidentiality and theStructure of Speech Acts. PhD thesis, Rutgers.

Palmer, F.R. 1987. Mood and Modality. Cambridge UP.Portner, Paul 1998. The Progressive in Modal Semantics. Language 74(4): 760-87Portner, P. 2009. Modality. Oxford University Press.Prior, A. 1957. Time and Modality. Oxford: Oxford University Press.Smirnova, A. 2012. Mood in Bulgarian. To appear in CLS 48.Squartini, M. 2004. Disentangling evidentiality and epistemic modality in Romance.

Lingua, 114 : 873-895.Thomason, R. 1984. Combination of Tense and Modality, in D. Gabbay; F. Guenthner

(eds.), Handbook of Philosophical Logic: Extensions of Classical Logic, Dordrecht: Reidel,135-165.

Tsangalidis, Anastasios. 1998. Will and tha: a Comparative Study of the CategoryFuture. Thessaloniki: University Studio Press.

22